FSD V14 Lite — What 26 Korean Owners Scored It, and Why

What is FSD V14 Lite, and why "Lite"?

FSD stands for Full Self-Driving, the name of Tesla's driver-assistance software. Despite the name, the actual tier is Supervised: the driver must keep watching the road and remains responsible for intervening. That gap between the name and the tier is the first thing that trips up newcomers. The accurate description is "the car does the operating and the person supervises".

The "Lite" comes from hardware. The self-driving compute chip in Tesla vehicles comes in generations — the one used from 2019 is called HW3, later ones HW4. HW3 has roughly 15% of HW4's memory bandwidth, so the latest FSD cannot be loaded onto it as is. V14 Lite is the version cut down to fit those HW3 cars.

The reviews covered here come from owners actually running V14 Lite in Korea. The raw performance gap between hardware generations was covered in an earlier piece that drove the same route on both; this one is not a performance comparison but a record of user reactions.

Where these scores came from

Tslachan, a Korean YouTube channel covering Tesla, posted on its X account asking anyone who had experienced FSD V14 Lite in Korea to leave a score along with an honest account. The question was simple: out of 100, what would you give it?

Twenty-six people answered with a number. Nine of them said 90 — about 35%, the most common answer. The highest was a single 100, the lowest three 70s. Overall most responses sat above 80, with the nineties accounting for close to half.

The channel says the same thing twice, at the start and at the end: the sample is far too small to represent FSD Lite in any absolute sense, and it should be taken as interest and reference only. That caveat has to travel with the numbers wherever they are quoted. Twenty-six replies are less a statistic than a window onto what the people currently using this feature are saying.

What owners liked

The defining feature of the positive comments is that they do not point at anything specific. Rather than naming a stretch of road where the car did something well, owners said the driving process as a whole was excellent and satisfying. Several reported covering the distance with almost no intervention apart from navigation errors.

The most quotable reaction came second-hand, from one owner's wife: it drives better than you do. Coming from the passenger seat rather than a spec sheet, it is an oblique way of saying how natural the driving feels.

There was admiration on the hardware side too. One owner called it remarkable that a seven-year-old chip in a five-year-old car gained autonomous driving through a single software update. Getting a capability the car did not have yesterday, without buying a new one, is still an unfamiliar experience in automobiles.

And one word recurs across the replies: release. The sense of being freed from the labour of driving, and a novelty that has not worn off, dominate the responses.

What owners found lacking

The complaints have the opposite shape. They do not generalise; each one specifies what fails and under what conditions. Below is the list as it appeared in the replies, paired with the situation it occurs in.

Navigation came up most often. Owners said the car does not take the route they want, making the daily commute awkward, and that at critical points where lanes divide it drives differently from what the map says. Worth noting: this is not a driving-skill problem but a route-judgement problem. The car drives well — it just drives well toward the wrong place.

Lane changes drew criticism too: unnecessary changes happen often, including drifting toward bus and taxi stops. Signal recognition was weakest at night. Left-turn signals in particular are read poorly, so the car fails to register that the light has changed and sits still. The driver ends up feathering the accelerator.

On speed and driving style the complaints ran in both directions. Some found it frustrating that the car will not exceed the posted limit on the motorway; others reported the car reading the speed limit as lower where navigation and GPS accuracy are poor, causing an abrupt drop in speed. Beyond that: phantom braking before speed bumps, needing care when passing through Hi-pass toll gates, and recognising a speed camera without slowing for it.

Why praise generalises and complaints have addresses

The same 26 replies describe the two directions at different resolutions. What went well stops at "overall"; what went badly goes down to "at night", "at the fork", "before the speed bump". The asymmetry is itself information.

Actions that succeed do not register. Holding a lane, keeping distance, matching speed at a merge — when these keep working they stop being events and become background. Failures, by contrast, are events every time. If the light changes and the car sits there, the driver remembers that moment and tenses up in advance at the same junction next time.

So the shape of this feedback does not mean the feature is bad. It also means success has become stable enough to sink into the background. And it shows that the remaining failures are not random but concentrated in a few narrow situations: night signals, route judgement, road surface irregularities. They can be written down as a list.

That observation is useful when reading the scores too. A 90 does not mean it fails once in ten. It means most of the time is background, and a person steps in during particular situations.

What changes on 21 August

One comment among the replies may outlast the scores. Since what counts as 100 is subjective and everyone's circumstances differ, the writer suggested treating "is it worth paying a monthly subscription for" as the reference point rather than any absolute score.

That comment has a regulatory backdrop. On 7 August 2026, Tesla Korea announced a change to how FSD and EAP (Enhanced Autopilot) are sold. The start date, originally 10 August, was pushed to the 21st.

The changes are below. Anyone who has already bought outright is unaffected and keeps what they have.

A score may reflect a free trial; a subscription comes out of the wallet every month. So the verdicts from people who actually subscribe after the 21st are likely to be considerably colder than these 26. That is a good reason to run the same question again later.

How to read these numbers

Twenty-six is too few to use as a statistic, exactly as the channel stressed twice. On top of that, the respondents follow a particular Tesla channel on X and cared enough to comment, so they lean toward interest in and goodwill toward the feature to begin with. Quoting the average score elsewhere is best avoided.

Something usable remains, though: not the height of the scores but the list of reasons points came off. That eight specific complaints recurred across different people means they are reproducible situations rather than individual bad luck.

So here is how to actually use these reviews. If you are weighing a subscription, count how many of the situations in the table above appear on your own routes. If your commute is full of forks and night-time left turns, the same software will feel lower-scoring to you; if it is mostly long motorway runs, the opposite. That calculation about your own roads predicts your experience far better than an average of 26 people.